Separate and Unequal

ORIGINAL INVESTIGATION

Separate and Unequal Clinics Where Minority and Nonminority Patients Receive Primary Care Anita B. Varkey, MD; Linda Baier Manwell, MS; Eric S. Williams, PhD; Said A. Ibrahim, MD, MPH; Roger L. Brown, PhD; James A. Bobula, PhD; Barbara A. Horner-Ibler, MD, MASW; Mark D. Schwartz, MD; Thomas R. Konrad, PhD; Jacqueline C. Wiltshire, PhD; Mark Linzer, MD; for the MEMO Investigators

Background: Few studies have examined the influ-

ence of physician workplace conditions on health care disparities. We compared 96 primary care clinics in New York, New York, and in the upper Midwest serving various proportions of minority patients to determine differences in workplace organizational characteristics. Methods: Cross-sectional data are from surveys of 96 clinic managers, 388 primary care physicians, and 1701 of their adult patients with hypertension, diabetes mellitus, or congestive heart failure participating in the Minimizing Error, Maximizing Outcome (MEMO) study. Data from 27 clinics with at least 30% minority patients were contrasted with data from 69 clinics with less than 30% minority patients. Results: Compared with clinics serving less than 30%

minority patients, clinics serving at least 30% minority patients have less access to medical supplies (2.7 vs 3.4, P⬍ .001), referral specialists (3.0 vs 3.5, P ⬍.005) on a scale of 1 (none) to 4 (great), and examination rooms per physician (2.2 vs 2.7, P=.002) . Their patients are more

frequently depressed (22.8% vs 12.1%), are more often covered by Medicaid (30.2% vs 11.4%), and report lower health literacy (3.7 vs 4.4) on a scale of 1 (low) to 5 (high) (P ⬍ .001 for all). Physicians from clinics serving higher proportions of minority populations perceive their patients as frequently speaking little or no English (27.1% vs 3.4%, P =.004), having more chronic pain (24.1% vs 12.9%, P⬍ .001) and substance abuse problems (15.1% vs 10.1%, P=.005), and being more medically complex (53.1% vs 39.9%) and psychosocially complex (44.9% vs 28.2%) (P⬍.001 for both). In regression analyses, clinics with at least 30% minority patients are more likely to have chaotic work environments (odds ratio, 4.0; P=.003) and to have fewer physicians reporting high work control (0.2; P =.003) or high job satisfaction (0.4; P=.01). Conclusion: Clinics serving higher proportions of minority patients have more challenging workplace and organizational characteristics.

Arch Intern Med. 2009;169(3):243-250

M

Author Affiliations are listed at the end of this article. Group Information: Investigators in the Minimizing Error, Maximizing Outcome (MEMO) study are listed at the end of the article.

INORITY A MERICANS have poorer health outcomes from chronic conditions such as cancer, asthma, heart disease, and diabetes mellitus.1 Some of these poorer outcomes are attributable to disparities in health care resulting from a myriad of access, patient, and physician factors.1 Potential patient-related factors include trust, literacy, attitudes, education, knowledge, preferences, health beliefs, cultural traditions, late-stage presentation of illness, and racial/ethnic concordance with physicians.2 Disparities can also result from insurance status, health care affordability, inadequate access to care and transportation, and competing demands, such as employment and child care.2 Finally, physician factors, including bias, poor communication skills, and inadequate training in cross-cultural issues, may also be involved.2

(REPRINTED) ARCH INTERN MED/ VOL 169 (NO. 3), FEB 9, 2009 243

Bach et al3 investigated if disparities in care could be explained by differences in access to resources. They found that a limited number of physicians provided most of the care for African Americans and that physicians caring for these patients reported limited access to health care resources, such as specialists and diagnostic imaging. However, that study provided little information about the work environment of physicians providing care to minority patients. This study extends the work of Bach et al3 by comparing the workplace characteristics of primary care clinics having sizeable minority clienteles with those of primary care clinics having mostly nonminority clienteles. We hypothesized that clinics serving more minority patients would have a more complex patient mix, increased adverse outcomes among physicians, and greater challenges in workplace characteristics.

WWW.ARCHINTERNMED.COM

Downloaded from www.archinternmed.com at Health Services Research Library, on March 20, 2009 ©2009 American Medical Association. All rights reserved.

Table 1. Structure of the Minimizing Error, Maximizing Outcome (MEMO) Study Clinics No. (%) Clinics Serving ⱖ30% Minority Other P Patients Clinics (n=27) (n=69) Value a

Variable Clinic type Academic affiliation Community primary care–only clinic Community multispecialty clinic Hospital-based primary care clinic Clinic financing Health maintenance organization County Hospital University Physician owned a Adjusted

10 (37.0) 14 (51.9)

31 (44.9) 32 (46.4)

.69 .73

10 (37.0) 3 (11.1)

35 (50.7) 2 (2.9)

.36 .25

6 (22.2) 9 (33.3) 5 (18.5) 5 (18.5) 2 (7.4)

33 (47.8) .11 1 (1.4) ⬍.001 3 (4.3) .14 27 (39.1) .14 5 (7.2) .99

for multiplicity of tests (false discovery rates).

METHODS

SUBJECTS AND STUDY DESIGN Patients, physicians, and clinic managers from 119 primary care clinics participated in the Minimizing Error, Maximizing Outcome (MEMO) study. This 4-year (2001-2005) multimethod longitudinalinvestigationassessedhowhealthcareworkplacefactorsaffect the quality of medical care.4 The 118 practices were located in 5 regions, including inner-city clinics in New York, New York, and Chicago, Illinois; academic and managed care clinics in Milwaukee and Madison, Wisconsin; and small town or rural private practice clinics in central Wisconsin. Practices in these areas were solicited for their diverse patient base, wide range of payers, and high proportions of uninsured patients. We emphasized recruitment of clinics serving large numbers of minority patients to address special issues for these populations. The institutional review board at each participating organization reviewed and approved the researchprotocol,andallparticipants(patients,physicians,andclinic managers) provided written consent. A physician or PhD-level site director in each region facilitated the recruitment process. Eligible physician participants were family practitioners or general internists who spent at least 4 sessions per week providing ambulatory primary care. Clinics were enrolled if at least 50% of their physicians chose to participate. Up to 6 patients for each participating physician were surveyed by local MEMO study researchers via mail or waiting room recruitment. Patient eligibility criteria included age 18 years or older; ability to read in English, Spanish, or Cantonese; at least 2 outpatient visits with a participating physician in the previous 12 months; and a diagnosis of hypertension, diabetes mellitus, or congestive heart failure. The patient surveys and consent forms were created in English, translated into Spanish and Cantonese, and independently backtranslated by bilingual professional translators.

MAIN MEASUREMENTS This article reports data from surveys of patients, physicians, and clinic managers. The physician survey was derived from the Physician Worklife Survey,5 supplemented by physician comments from focus groups conducted at the inception of the MEMO study. The physician survey included a single query regarding burn-

out6 and a question about the physician’s likelihood of leaving the practice within 2 years. Physicians also rated the pace of the office on a 5-point scale ranging from calm to chaotic; clinics with a mean score of 4 or higher were deemed chaotic. Organizational climate was assessed using a modified version of the multidimensional measure by Kralewski et al.7 To decrease respondent burden and to ameliorate overlap with other measures, we tested 6 of 9 domains pertinent to the present study as follows: collegiality, cohesiveness, organizational trust, quality emphasis, information emphasis, and organizational identity. Split-sample exploratory and confirmatory factor analysis of the modified measure revealed the following 5-factor structure: (1) alignment between leadership and physician values (8 items, ␣=.86), (2) practice emphasis on quality (6 items, ␣=.88), (3) sense of trust or belonging (5 items, ␣=.79), (4) practice emphasis on information and communication (4 items, ␣=.70), and (5) cohesiveness (3 items, ␣=.66). The values alignment scale was new and was not found in the study by Kralewski et al. Physicians also provided information about age, sex, marital status, medical specialty, and income range. In compliance with the 1997 Office of Management and Budget standards,8 they also responded to race/ethnicity queries that included 2 ethnic categories (Hispanic or Latino vs not Hispanic or Latino) and 5 racial categories (White, Asian, black or African American, American IndianorAlaskaNative,andNativeHawaiianorOtherPacificIslander). The organization assessment survey was generated from physician investigators’ personal experience and from comments of focus groups convened at the inception of the MEMO study. Clinic managers provided information about clinic structure (eg, payer and patient mix), processes (eg, electronic medical records and bottlenecks), and management (eg, staff meetings and quality management). Clinic managers also provided race/ ethnicity and payer mix information. The patient survey, based in part on comments elicited during patient focus groups,9 queried about satisfaction with physicians and clinics,10 health literacy,11 trust in the physician,12 overall and disease-related quality of life,13,14 and symptoms of depression.15 Patients also provided information about marital status and educational level. Information about race/ethnicity was provided per Office of Management and Budget standards.8

STATISTICAL ANALYSIS To compare clinics that care for large numbers of minority patients with those that do not, we chose a cut point of 30%. This threshold was selected because it is similar to the proportion of African Americans plus Hispanics in the US population according to 2000 census data.16 Exact 2-binomial unconditional tests, parametric and nonparametric mean contrasts (where appropriate), and 2-level logistic regression analyses were used to determine differences between the 2 groups of clinics. Two adjustments were applied to our analyses. First, adjustment was made to our clustered data standard error estimates becausetraditionalestimatescanprovidenegativelybiasedestimates.17 For example, physician data nested under organizations and patient data nested under physicians are more likely to be correlated with each other. Therefore, we used the Huber-White sandwich estimator to correct the negatively biased standard errors.18 Second, because this analysis incorporates multiple statistical assessments, the chance of making type I errors is increased, and our power for an individual test may become unacceptably low. Instead of highly conservative familywise error rate adjustments (eg, Bonferroni, Sidak, and others), we used the false discovery rate approach to multiple hypothesis testing. The false discovery rate approach controls the expected proportion of incorrectly rejected null hypotheses (type I errors) among all the rejected hypotheses. It provides a good balance between discovery of statistically significant effects and limitation of false-positive occurrences.19

(REPRINTED) ARCH INTERN MED/ VOL 169 (NO. 3), FEB 9, 2009 244

WWW.ARCHINTERNMED.COM

Downloaded from www.archinternmed.com at Health Services Research Library, on March 20, 2009 ©2009 American Medical Association. All rights reserved.

Table 2. Physician and Patient Characteristics in Clinics Serving Larger Proportions of Minority Patients vs Clinics That Do Not a

Variable

Clinics Serving ⱖ30% Minority Patients (n = 27)

Other Clinics (n = 69)

(n = 162) 115 (71.0) 47 (29.0) 84 (51.9) 41 (10) 64 (39.5) (n = 780) 585 (75.0) 339 (43.5) 29 (3.7) 99 (12.7) 199 (25.5) 424 (54.4)

(n = 226) 90 (39.8) 136 (60.2) 93 (41.2) 44 (9) 25 (11.1) (n = 921) 85 (9.2) 49 (5.3) 11 (1.2) 14 (1.5) 13 (1.4) 536 (58.2)

180 (23.1) 280 (35.9) 187 (24.0) 84 (10.8) 49 (6.3)

49 (5.3) 247 (26.8) 273 (29.6) 170 (18.5) 182 (19.8)

⬍.001 ⬍.001 .08 ⬍.001 ⬍.001

49 (41) 26 (19) 30 (20) 28 (32) 3.7 (1.0)

75 (33) 26 (15) 11 (8) 9 (13) 4.4 (0.8)

.02 .80 ⬍.001 .048 ⬍.001

701 (89.9) 445 (57.1) 162 (20.8) 358 (45.9)

774 (84.0) 516 (56.0) 186 (20.2) 436 (47.3)

⬍.001 .39 .31 .55

139 (19.8) 122 (27.4) 54 (33.3) 178 (22.8) 177 (22.7)

64 (8.3) 114 (22.1) 42 (22.6) 111 (12.1) 243 (26.4)

⬍.001 .07 .04 ⬍.001 .09

Physicians General internists, No. (%) Family practitioners, No. (%) Female, No. (%) Age, mean (SD) Racial/ethnic minority, No. (%) Patients Racial/ethnic minority, No. (%) c Black or African American Asian Other Hispanic or Latino Female, No. (%) Educational level, No. (%) ⱕHigh school High school graduate Some college College graduate Graduate or professional school Insurance status, %, mean (SD) d Insured Medicare Medicaid Uninsured or self-pay Medical literacy, mean (SD) e Presence of target disease, No. (%) Hypertension Diabetes mellitus Congestive heart failure ⬎1 Target disease Limited by disease always or very/fairly often for those who have the disease, No. (%) Patients with hypertension Patients with diabetes mellitus Patients with congestive heart failure Symptoms of depression (ⱖ7 d in past 2 wk), No. (%) Taking antidepressant medication, No. (%)

a Data are from patient, physician, and clinic manager surveys. b Adjusted for multiple comparisons and dependency (lack of independence of patients clustered c Racial/ethnic categories are not mutually exclusive. d Subcategories do not sum to 100.0% because of aggregate central tendency measures. e On a scale of 1 (low) to 5 (high).

RESULTS

PHYSICIAN AND PATIENT CHARACTERISTICS A total of 443 physicians (58.8% of those approached) consented to participate in the MEMO study, and 422 (95.3%) of these completed the survey. Nonparticipants did not differ substantially in specialty or sex from physicians who chose to participate. The final enrollment of 422 represents 84.4% of our original target sample of 500 physicians. Physician participants practiced in 118 primary care clinics with a diverse patient base, wide range of payers, and large numbers of indigent or uninsured patients. Of 135 clinics where physicians were interested in participating, 118 clinics met the criteria of at least 50% physician participation. Clinics that did not meet the criteria were likely to be smaller, with fewer physicians in the practice. A total of 1795 patients completed the patient survey, a mean of 4 patients per participat-

P Value b

⬍.001 .03 ⬍.001 ⬍.001 ⬍.001 ⬍.001 ⬍.001 ⬍.001 ⬍.001 .03

under physicians and physicians nested within clinics).

ing physician. Because of institutional review board– mandated differences in patient solicitation methods, the participant rates varied by region and an overall rate could not be calculated. Waiting room solicitation was used in urban clinics because of a lack of computerized patient lists or missing or inaccurate patient addresses. Academic, suburban, and small-town clinics most often opted for mailed invitations. The waiting room method resulted in slightly higher recruitment rates. This article reports data from 388 physicians and 1701 patients from 96 clinics for which we have complete patient, physician, and clinic manager data (Table 1). The 22 excluded clinics were mostly affiliated with health maintenance organizations but did not differ statistically in size or geographic location from the clinics with complete data. Characteristics of the physician and patient participants are summarized in Table 2. Data provided by clinic managers revealed that 27 of 96 clinics had a client base composed of at least 30% minority patients. The percentages

(REPRINTED) ARCH INTERN MED/ VOL 169 (NO. 3), FEB 9, 2009 245

WWW.ARCHINTERNMED.COM

Downloaded from www.archinternmed.com at Health Services Research Library, on March 20, 2009 ©2009 American Medical Association. All rights reserved.

35 Nonminority clinics Minority clinics

No. of Clinics

23

11

0.0 0.0

10.0

20.0

30.0

40.0

50.0

60.0

70.0

80.0

90.0 100.0

Minority Patients, %

Figure 1. Distribution of within-clinic percentages of minority patients across 96 clinics in the Minimizing Error, Maximizing Outcome (MEMO) study.

1

Minority classification Nonminority classification More GIM MDs∗

Scale

More female MDs∗

Less insured∗

More